[Bug target/116576] New: `INT64_MAX` and `-Oz`

[Bug target/116576] New: `INT64_MAX` and `-Oz`

[lh_mouse at 126 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116576

            Bug ID: 116576
           Summary: `INT64_MAX` and `-Oz`
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: lh_mouse at 126 dot com
  Target Milestone: ---

For this simple function:

   ```
   int64_t make_value(void) { return INT64_MAX; }
   ```

Both GCC and Clang generate an 8-byte `movabs` which adds up to 10 bytes:

   ```
   48 b8 ffffffffffffff7f    ; mov rax, 0x7fffffffffffffff
                             ; (10 bytes; no dependency)
   ```

When we do `-Oz` we otpimize for size and do not care about the false
dependency, and this can be accomplished in only 7 bytes:

   ```
   48 83 c8 ff               ; or rax, -1
   48 d1 e8                  ; shr rax
                             ; (7 bytes; input dependency on rax)
   ```

[Bug target/116576] `INT64_MAX` and `-Oz`

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116576

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement

[Bug target/116576] `INT64_MAX` and `-Oz`

[lh_mouse at 126 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116576

--- Comment #1 from LIU Hao <lh_mouse at 126 dot com> ---
Here is a longer solution but it breaks the dependency:

   ```
   31c0            ; xor eax, eax
   48 ffc8         ; dec rax
   48 d1e8         ; shr rax
                   ; (8 bytes; no dependency)
   ```

[Bug target/116576] `INT64_MAX` and `-Oz`

[roger at nextmovesoftware dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116576

Roger Sayle <roger at nextmovesoftware dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
                 CC|                            |roger at nextmovesoftware dot com
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-09-09

--- Comment #2 from Roger Sayle <roger at nextmovesoftware dot com> ---
The constant ~0 can be materialized on x86 in only three bytes using either of
the sequences "push -1; pop rax" (i.e. 6a ff 58) or "stc; sbb ax,ax" (i.e. f9
19 c0), both of which could theoretically be fused/optimized/avoid dependencies
on modern hardware.  With -Oz, GCC mainline and clang currently use the first
sequence.

But you're absolutely right that following this by a logical shift can be used
to materialize INT64_MAX (and other constants) in fewer bytes than the standard
movabs when its safe to clobber the flags.